Public/koboldcpp
1
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-09-18 16:55:14 +02:00
Code Issues Packages Projects Releases Wiki Activity

rename tokens tab to context, move fa to hardware

Browse Source
This commit is contained in:
Concedo
2025-12-26 00:06:07 +08:00
parent 062f8b28eb
commit 399fc9c57e
1 changed files with 30 additions and 29 deletions
Download Patch File Download Diff File Expand all files Collapse all files
koboldcpp.py
+30 -29
View File
@@ -5155,7 +5155,7 @@ def show_gui():
tabs = ctk.CTkFrame(root, corner_radius = 0, width=windowwidth, height=windowheight-50)
tabs.grid(row=0, stick="nsew")
tabnames= ["Quick Launch", "Hardware", "Tokens", "Loaded Files", "Network", "Horde Worker","Image Gen","Audio","Admin","Extra"]
tabnames= ["Quick Launch", "Hardware", "Context", "Loaded Files", "Network", "Horde Worker","Image Gen","Audio","Admin","Extra"]
navbuttons = {}
navbuttonframe = ctk.CTkFrame(tabs, width=int(104), height=int(tabs.cget("height")))
navbuttonframe.grid(row=0, column=0, padx=2,pady=2)
@@ -5894,28 +5894,30 @@ def show_gui():
makeslider(hardware_tab, "Batch Size:", batchsize_text, blas_size_var, 0, len(batchsize_values)-1, 16,width=200, set=6,tooltip="How many tokens to process at once per batch.\nLarger values use more memory.")
blas_size_var.trace_add("write", changed_gpulayers_estimate)
makecheckbox(hardware_tab, "AutoFit (llama.cpp mode)", autofit_var, 100,0, tooltiptxt="Automatically attempt to fit the model in the best possible way. Overrides everything else. Not recommended for multi model setups. Experimental.")
makecheckbox(hardware_tab, "Use FlashAttention", flashattention_var, 100, command=toggleflashattn, tooltiptxt="Enable flash attention for GGUF models.")
makecheckbox(hardware_tab, "AutoFit (llama.cpp mode)", autofit_var, 100,0,padx=160, tooltiptxt="Automatically attempt to fit the model in the best possible way. Overrides everything else. Not recommended for multi model setups. Experimental.")
ctk.CTkButton(hardware_tab , text = "Run Benchmark", command = guibench ).grid(row=110,column=0, stick="nw", padx= 8, pady=2)
# Tokens Tab
tokens_tab = tabcontent["Tokens"]
# tokens checkboxes
smartcontextbox = makecheckbox(tokens_tab, "Use SmartContext", smartcontext_var, 1,tooltiptxt="Uses SmartContext. Now considered outdated and not recommended.\nCheck the wiki for more info.")
makecheckbox(tokens_tab, "Use ContextShift", contextshift_var, 2,tooltiptxt="Uses Context Shifting to reduce reprocessing.\nRecommended. Check the wiki for more info.", command=togglectxshift)
makecheckbox(tokens_tab, "Use FastForwarding", fastforward_var, 3,tooltiptxt="Use fast forwarding to recycle previous context (always reprocess if disabled).\nRecommended.", command=togglefastforward)
makecheckbox(tokens_tab, "Use Sliding Window Attention (SWA)", swa_var, 4,tooltiptxt="Allows Sliding Window Attention (SWA) KV Cache, which saves memory but cannot be used with context shifting.", command=toggleswa)
makecheckbox(tokens_tab, "Use SmartCache", smartcache_var, 5,tooltiptxt="Enables intelligent context switching by saving KV cache snapshots to RAM. Requires fast forwarding.", command=togglesmartcache)
# Context Tab
context_tab = tabcontent["Context"]
# Context checkboxes
smartcontextbox = makecheckbox(context_tab, "Use SmartContext", smartcontext_var, 1,tooltiptxt="Uses SmartContext. Now considered outdated and not recommended.\nCheck the wiki for more info.")
makecheckbox(context_tab, "Use ContextShift", contextshift_var, 2,tooltiptxt="Uses Context Shifting to reduce reprocessing.\nRecommended. Check the wiki for more info.", command=togglectxshift)
makecheckbox(context_tab, "Use FastForwarding", fastforward_var, 3,tooltiptxt="Use fast forwarding to recycle previous context (always reprocess if disabled).\nRecommended.", command=togglefastforward)
makecheckbox(context_tab, "Use Sliding Window Attention (SWA)", swa_var, 4,tooltiptxt="Allows Sliding Window Attention (SWA) KV Cache, which saves memory but cannot be used with context shifting.", command=toggleswa)
makecheckbox(context_tab, "Use SmartCache", smartcache_var, 5,tooltiptxt="Enables intelligent context switching by saving KV cache snapshots to RAM. Requires fast forwarding.", command=togglesmartcache)
# context size
makeslider(tokens_tab, "Context Size:",contextsize_text, context_var, 0, len(contextsize_text)-1, 18, width=280, set=7,tooltip="What is the maximum context size to support. Model specific. You cannot exceed it.\nLarger contexts require more memory, and not all models support it.")
makeslider(context_tab, "Context Size:",contextsize_text, context_var, 0, len(contextsize_text)-1, 18, width=280, set=7,tooltip="What is the maximum context size to support. Model specific. You cannot exceed it.\nLarger contexts require more memory, and not all models support it.")
context_var.trace_add("write", changed_gpulayers_estimate)
makelabelentry(tokens_tab, "Default Gen Amt:", defaultgenamt_var, row=20, padx=(120), singleline=True, tooltip="How many tokens to generate by default, if not specified. Must be smaller than context size. Usually, your frontend GUI will override this.")
makelabelentry(tokens_tab, "Prompt Limit:", genlimit_var, row=20, padx=(300), singleline=True, tooltip="If set, restricts max output tokens to this limit regardless of API request. Set to 0 to disable.",labelpadx=(210))
makelabelentry(context_tab, "Default Gen Amt:", defaultgenamt_var, row=20, padx=(120), singleline=True, tooltip="How many tokens to generate by default, if not specified. Must be smaller than context size. Usually, your frontend GUI will override this.")
makelabelentry(context_tab, "Prompt Limit:", genlimit_var, row=20, padx=(300), singleline=True, tooltip="If set, restricts max output tokens to this limit regardless of API request. Set to 0 to disable.",labelpadx=(210))
nativectx_entry, nativectx_label = makelabelentry(tokens_tab, "Override Native Context:", customrope_nativectx, row=23, padx=(146), singleline=True, tooltip="Overrides the native trained context of the loaded model with a custom value to be used for Rope scaling.")
customrope_scale_entry, customrope_scale_label = makelabelentry(tokens_tab, "RoPE Scale:", customrope_scale, row=23, padx=(100), singleline=True, tooltip="For Linear RoPE scaling. RoPE frequency scale.")
customrope_base_entry, customrope_base_label = makelabelentry(tokens_tab, "Base:", customrope_base, row=23, padx=(220), singleline=True, tooltip="For NTK Aware Scaling. RoPE frequency base.",labelpadx=(180))
nativectx_entry, nativectx_label = makelabelentry(context_tab, "Override Native Context:", customrope_nativectx, row=23, padx=(146), singleline=True, tooltip="Overrides the native trained context of the loaded model with a custom value to be used for Rope scaling.")
customrope_scale_entry, customrope_scale_label = makelabelentry(context_tab, "RoPE Scale:", customrope_scale, row=23, padx=(100), singleline=True, tooltip="For Linear RoPE scaling. RoPE frequency scale.")
customrope_base_entry, customrope_base_label = makelabelentry(context_tab, "Base:", customrope_base, row=23, padx=(220), singleline=True, tooltip="For NTK Aware Scaling. RoPE frequency base.",labelpadx=(180))
def togglerope(a,b,c):
manualropebox.grid_remove()
nativectx_label.grid_remove()
@@ -5935,16 +5937,15 @@ def show_gui():
nativectx_label.grid(row=23, column=0, padx=8, pady=1, stick="nw")
nativectx_entry.grid(row=23, column=0, padx=(146), pady=1, stick="nw")
manualropebox = makecheckbox(tokens_tab, "Manual Rope Scale", variable=manualrope_var, row=22, command=togglerope, padx=(200), tooltiptxt="Set RoPE base and scale manually.")
manualropebox = makecheckbox(context_tab, "Manual Rope Scale", variable=manualrope_var, row=22, command=togglerope, padx=(200), tooltiptxt="Set RoPE base and scale manually.")
makecheckbox(tokens_tab, "Custom RoPE Config", variable=customrope_var, row=22, command=togglerope,tooltiptxt="Override the default RoPE configuration with custom RoPE scaling.")
makecheckbox(tokens_tab, "Use FlashAttention", flashattention_var, 28, command=toggleflashattn, tooltiptxt="Enable flash attention for GGUF models.")
noqkvlabel = makelabel(tokens_tab,"(Note: QuantKV works best with flash attention)",28,0,"Only K cache can be quantized, and performance can suffer.\nIn some cases, it might even use more VRAM when doing a full offload.",padx=160)
makecheckbox(context_tab, "Custom RoPE Config", variable=customrope_var, row=22, command=togglerope,tooltiptxt="Override the default RoPE configuration with custom RoPE scaling.")
noqkvlabel = makelabel(context_tab,"(Note: QuantKV works best with flash attention)",30,0,"Only K cache can be quantized, and performance can suffer.\nIn some cases, it might even use more VRAM when doing a full offload.",padx=160)
noqkvlabel.configure(text_color="#ff5555")
qkvslider,qkvlabel,qkvtitle = makeslider(tokens_tab, "Quantize KV Cache:", quantkv_text, quantkv_var, 0, 2, 30, set=0,tooltip="Enable quantization of KV cache.\nRequires FlashAttention for full effect, otherwise only K cache is quantized.")
qkvslider,qkvlabel,qkvtitle = makeslider(context_tab, "Quantize KV Cache:", quantkv_text, quantkv_var, 0, 2, 30, set=0,tooltip="Enable quantization of KV cache.\nRequires FlashAttention for full effect, otherwise only K cache is quantized.")
quantkv_var.trace_add("write", toggleflashattn)
makecheckbox(tokens_tab, "No BOS Token", nobostoken_var, 43, tooltiptxt="Prevents BOS token from being added at the start of any prompt. Usually NOT recommended for most models.")
makecheckbox(tokens_tab, "Enable Guidance", enableguidance_var, 43,padx=(140), tooltiptxt="Enables the use of Classifier-Free-Guidance, which allows the use of negative prompts. Has performance and memory impact.")
makecheckbox(context_tab, "No BOS Token", nobostoken_var, 43, tooltiptxt="Prevents BOS token from being added at the start of any prompt. Usually NOT recommended for most models.")
makecheckbox(context_tab, "Enable Guidance", enableguidance_var, 43,padx=(140), tooltiptxt="Enables the use of Classifier-Free-Guidance, which allows the use of negative prompts. Has performance and memory impact.")
def togglejinja(a,b,c):
if jinja_var.get()==1:
jinjatoolsbox.grid()
@@ -5952,13 +5953,13 @@ def show_gui():
jinja_tools_var.set(0)
jinjatoolsbox.grid_remove()
changed_gpulayers_estimate()
makecheckbox(tokens_tab, "Use Jinja", jinja_var, row=45, command=togglejinja, tooltiptxt="Enables using jinja chat template formatting for chat completions endpoint. Other endpoints are unaffected.")
jinjatoolsbox = makecheckbox(tokens_tab, "Jinja for Tools", jinja_tools_var, row=45 ,padx=(140), tooltiptxt="Allows jinja even with tool calls. If unchecked, jinja will be disabled when tools are used.")
makecheckbox(context_tab, "Use Jinja", jinja_var, row=45, command=togglejinja, tooltiptxt="Enables using jinja chat template formatting for chat completions endpoint. Other endpoints are unaffected.")
jinjatoolsbox = makecheckbox(context_tab, "Jinja for Tools", jinja_tools_var, row=45 ,padx=(140), tooltiptxt="Allows jinja even with tool calls. If unchecked, jinja will be disabled when tools are used.")
jinja_var.trace_add("write", togglejinja)
makelabelentry(tokens_tab, "MoE Experts:", moeexperts_var, row=55, padx=(120), singleline=True, tooltip="Override number of MoE experts.")
makelabelentry(tokens_tab, "MoE CPU Layers:", moecpu_var, row=55, padx=(320), singleline=True, tooltip="Force Mixture of Experts (MoE) weights of the first N layers to the CPU.\nSetting it higher than GPU layers has no effect.", labelpadx=(210))
makelabelentry(tokens_tab, "Override KV:", override_kv_var, row=57, padx=(120), singleline=True, width=150, tooltip="Override metadata value by key. Separate multiple values with commas. Format is name=type:value. Types: int, float, bool, str")
makelabelentry(tokens_tab, "Override Tensors:", override_tensors_var, row=59, padx=(120), singleline=True, width=150, tooltip="Override selected backend for specific tensors matching tensor_name_regex_pattern=buffer_type, same as in llama.cpp.")
makelabelentry(context_tab, "MoE Experts:", moeexperts_var, row=55, padx=(120), singleline=True, tooltip="Override number of MoE experts.")
makelabelentry(context_tab, "MoE CPU Layers:", moecpu_var, row=55, padx=(320), singleline=True, tooltip="Force Mixture of Experts (MoE) weights of the first N layers to the CPU.\nSetting it higher than GPU layers has no effect.", labelpadx=(210))
makelabelentry(context_tab, "Override KV:", override_kv_var, row=57, padx=(120), singleline=True, width=150, tooltip="Override metadata value by key. Separate multiple values with commas. Format is name=type:value. Types: int, float, bool, str")
makelabelentry(context_tab, "Override Tensors:", override_tensors_var, row=59, padx=(120), singleline=True, width=150, tooltip="Override selected backend for specific tensors matching tensor_name_regex_pattern=buffer_type, same as in llama.cpp.")
# Model Tab
model_tab = tabcontent["Loaded Files"]